Agents / Support
An AI support agent that answers from your knowledge base, and knows when to stop
Repetitive tickets bury the ones that need a human.
What it does
- Triages and tags incoming tickets
- Drafts replies grounded in your help centre
- Escalates with a summary and full context
- Never invents policies, refunds or promises
What you get
3 weeks. Fixed price from €7,500, ex. VAT.
The agent, in production
Running on your stack, on one process.
A test report
What we attacked, what broke and how we fixed it.
Documentation
Workflows, prompts and how to change them.
30 days of support
After launch, while your team gets used to it.
Red team
How it's tested
Jailbreaks, prompt injection through tickets and attachments, attempts to extract the system prompt or other customers' data, and pressure to promise refunds. We run them all before launch.
About our security reviews- Jailbreaks
- Prompt injection through tickets and attachments
- Extracting the system prompt
- Extracting other customers' data
- Pressure to promise refunds
How it works
- 01
Pick one process
A 20-minute call to find the work that eats the most hours, and to agree what success looks like.
- 02
Build and break it
Three weeks on your stack. Before launch we attack the agent and fix what breaks.
- 03
Hand over
Full documentation and a test report. Optional monthly retainer to keep improving it.
Questions
- Which tools do you work with?
- The ones you already use, including Zendesk, Intercom, Slack, n8n and Make.
- Which AI models?
- Claude, OpenAI or your approved provider. EU hosting if you need it.
- Who owns it?
- You do. The workflows, the prompts and the documentation are yours.
- What if it doesn't work?
- We agree the success criterion before we start.
What would you hand to an agent tomorrow?
One process, three weeks, from €7,500 ex. VAT.